Arxiv:1104.2854V1 [Q-Bio.PE] 14 Apr 2011

The contribution of statistical physics to evolutionary biology Harold P. de Vladara,∗, Nicholas H. Bartona,b aI.S.T. Austria. Klosterneuburg A-3400, Austria bInstitute of Evolutionary Biology, University of Edinburgh. Edinburgh, United Kingdom Abstract Evolutionary biology shares many concepts with statistical physics: both deal with populations, whether of molecules or organisms, and both seek to simplify evolution in very many dimensions. Often, methodologies have undergone parallel and independent development, as with stochastic methods in population genetics. We discuss aspects of population genetics that have embraced methods from physics: amongst others, non-equilibrium statistical mechanics, travelling waves, and Monte-Carlo methods have been used to study polygenic evolution, rates of adaptation, and range expansions. These applications indicate that evolutionary biology can further benefit from interactions with other areas of statistical physics, for example, by following the distribution of paths taken by a population through time. Keywords: Population Genetics, Statistical Thermodynamics, Evolutionary dynamics, Haldane’s Principle, Selection, Drift, Diffusion Equation, Fitness Flux, Entropy, Information, Travelling Waves, Monte Carlo. Parallel foundations of evolution and statistical physics In the late 19th century, Boltzmann established the theoretical foundations of statistical mechanics, in which the behaviour of ensembles of particles explains large-scale phenomena [1]. For example, the position and velocity of the particles in a gas can fluctuate between very many states (termed micro-states), but averages over all the configurations that give the same observ- able macroscopic states (temperature and pressure, say) [2]. A similar averaging over equivalent micro-states is made in both population and quantitative genetics: we average over individual gene combinations to describe a population by its allele frequencies, and we can further average over all the allele frequencies that are consistent with a given mean and variance of a quantitative trait. In this sense, physicists and evolutionary biologists both model populations (a gas or a gene pool) rather than precise types (individual particles or genotypes). This “statistical” arXiv:1104.2854v1 [q-bio.PE] 14 Apr 2011 description in terms of a few variables, the macro-states, summarizes the many possible configurations of the micro-states (degrees of freedom), which cannot be accurately measured or described. Furthermore, the macro-states are then sufficient to predict other properties without reference to the micro-states. For example, thermodynamics describes macroscopic properties ∗Corresponding author Email addresses: [email protected] (Harold P. de Vladar), [email protected] (Nicholas H. Barton) Preprint submitted to Trends in Ecology and Evolution May 22, 2018 without referring to individual particles; similarly, quantitative genetics does not refer to allele frequencies to predict the trait mean in the next generation. Hence, evolutionary biology and statistical physics often use similar theoretical methodologies, although studying very different phenomena. We argue that there are close analogies between evolutionary genetics and statistical physics. Physical techniques had an early influence on molecular biology (Appendix A). But more specifically, non-equilibrium methods are based on the same theory of stochastic processes that is used in population genetics. Thus, some physical theories promise further developments that can deepen our understanding of evolution in two ways: either by applying common mathematical techniques (e.g. diffusion equations, see Appendix B), or by developing precise analogies that incorporate new concepts (e.g. ensemble averaging, information, or entropy). These techniques are being applied in different aspects of evolutionary biology. This article focuses mainly in those that we consider most promising for population genetics. We aim to introduce to the reader to these methods by reviewing represen- tative examples in the literature. The cost of selection, entropy and information It is extraordinary that the selection of random mutations has created complex organisms that appear exquisitely designed to fit their environment. Selection can be seen as taking information from the environment, and coding it into the DNA sequence [3]: thus, the gene pool contains information about those specific sequences that confer high fitness. This idea can be quantified using the concepts of entropy and information [4]. Entropy is a measure of the number of different states in which a population is likely to be found: thus, selection of one specific genotype, or genotype frequency, corresponds to minimal entropy. A asking how the genotype of an individual, or a population, depends on the selection that they have experienced, can be quantified by an entropy that measures how strongly selection has clustered the population around a specific genotype. Haldane [5] showed that the number of selective deaths needed to fix an allele is independent of the selection pressure, and Kimura [6] pointed out that Haldane’s “cost of natural selection” is exactly the information gained by fixing a specific allele. This relation applies very generally to asexual populations [7] but fails with recombination (see below). The theory of quasispecies (a model of mutation-selection balance), emphasizes that the reproductive rate (selection) limits the amount of information that can be maintained in the face of random mutations [8]. However, this constraint can be relaxed with recombination and epistasis [9]. Analogies with statistical physics help us to understand how selection accumulates information. We first consider infinite populations – evolving deterministically – and then the more general case where random drift in finite populations drives evolution. Deterministic evolution and the role of recombination The information content of a single large population that is evolving deterministically is mea- P sured by the entropy, defined as S = x px log(px) where px is the frequency of alleles or genotype x [5, 8]. This entropy reflects the information accumulated and maintained by evolution, and is closely related to Shannon’s information [3-4]. Remarkably, the replicator dynamics can P d 2 be obtained by maximizing a different measure, Fisher information: F = x px dt log(px) , which measures divergence between two distributions [10, 11]. Fisher’s information is the “ac- 2 2 d celeration” of the entropy, i.e. d S=dt = F, where we interpret dt log(px) as the information contained about selection when we observe how the frequency of x has changed. For example, 2 for a beneficial allele under selection, this would be proportional to the selective value s. Nor- mally, we predict the change of the frequency distribution when we know s. Fisher’s information takes the parent and the offspring distributions as given, and measures the effect of selection from the difference between these two [11]. Sexual vs. asexual reproduction. It has long been understood that, when combined with trun- cation selection, sexual reproduction is much more efficient than asexual in fixing beneficial genotypes [12, 13, Ch. 2]. In the former case, the maximum information increases as n1=2 (n being the number of loci), while in the latter, it increases by only one unit per generation [3, 14]. The maximum number of loci that can be maintained despite the randomizing effect of mutation is 1/µ for asexuals [14], whilst for sexuals (with free recombination) it can be as high as 1/µ2, where µ is the mutation rate at each locus [14, 15]. (More loci would produce more mutants, and hence decrease the amount of information). According to Haldane’s principle [16] every delete- rious mutation must be eliminated by a failure to reproduce (a “selective death”). Therefore, the mutation load is independent of selection strength, and is half as great if selection eliminates two copies in a recessive homozygote at the same time. In haploids, redundancy leads to a similar gain in efficiency [15, 17]. Stochastic evolution: the diffusion of allele frequencies The diffusion approximation shows how the distribution of allele frequencies at many loci changes through time in finite populations. In this case, selection, mutation and migration are modelled as deterministic factors, and genetic drift introduces random fluctuations to populations within an ensemble (Appendix B). (Other treatments are possible, where mutations are regarded as carrying random changes to individuals within a single population.) In other fields, constant diffusion coefficients have been widely used, leading to simple Gaussian solutions (Appendix B). However, Gaussian solutions are not appropriate for population genetics, because allele frequencies range between zero and one, and sometimes cluster near fixation, in a bimodal distribution. After its introduction by Fisher [18], Kolmogorov applied the more general diffusion method to the neutral island model [19], which Wright [20] had already solved by different means. Kimura relied on the diffusion approximation to model the evolution of finite populations [21], and for his neutral theory of molecular evolution [22]. The diffusion approximation, central to both population genetics and statistical physics, pro- vides a way to model many factors in a mathematically tractable way. Crucially, it approximates a wide variety of more detailed models. Mathematically, it is equivalent to the coalescent pro- cess that

Load more